In recent years, stereo vision technology has gradually become a research focus in the field of computer vision technology and aims at acquiring depth images of an object through cameras. A depth image is an image reflecting the depth relationship of objects in a certain space, and in the image, pixel gray values represent the depth information in the space, namely the distance between points in the scene and the cameras. The depth image can be widely applied in three-dimensional reconstruction, collision detection, gesture recognition, robot navigation, design modeling for virtual scenes in movies and games, etc.
Currently, there are mainly two following methods to acquire a depth image. The first method is to acquire depth information of each point in the scene by direct observation via a measuring instrument (e.g., a camera with a function of measuring a distance). The second method is to acquire a depth image by calculation via a stereo matching method, and this method is to restore depth information of an object in a scene by the stereo matching of two parallax images of the same scene, acquired from two different viewpoints.